readf.x, readffn.x

Description

readf and readffn are TurboRVB utilities that read energy (or Forward Walking quantities) from fort.12 or fort.12.fn, form bins of length lbin, compute weighted bin averages and Jackknife error bars, and write fort.20 (averages and errors) and fort.21 (per-bin values). Both use the same program name erread in the source.

readf

Reads fort.12 (or fort.12.fn when iskipr < 0). Each record gives wbuf, wtot, optional eskip, and ebuf (energy). Bins of length lbin are formed; bins before ibinit are discarded. For each k = 0, …, maxk (bias correction factor), a different weight is applied (see Notes). Jackknife over the nbin bins yields gapjack (average) and gaps (standard error), written to fort.20. fort.21 gets one line per bin (k = maxk value, weight, bin index).

• Input: fort.12 or fort.12.fn; stdin: maxk, lbin, ibinitr, iskipr. Optional: command-line k-point index and kp_info.dat.

• Output: fort.20 (Independent bins, then “Energy , error, # of bias correcting factor” with gapjack, gaps for each k). fort.21 (per-bin line).

• Typical use: Post-process fort.12 from VMC or optimization to get binned energy average and error bar.

readffn

Extends readf by reading parminimized.d (optbra, tbra, etry, nsite) and supporting fort.12.fn format with pip (and pip0, pip1 when iskip = 0) for Forward Walking. When iskip = 0 and tbra ≠ 0, fort.20 writes “Growth estimate (H)” using \(E_{\mathrm{try}} - \ln(\mathtt{gapjack})/t_{\mathrm{bra}}\). optbra = -2 sets nodiv (alternative weight accumulation).

• Input: parminimized.d (required), fort.12 or fort.12.fn; stdin same as readf. Optional k-point and kp_info.dat.

• Output: fort.20 (Growth estimate when iskip=0 and tbra≠0; otherwise same as readf). fort.21 same as readf.

• Typical use: Post-process fort.12.fn from readforward.x to get energy or Growth estimate with Jackknife error.

Command line (both): First argument --help / -help / help — show help and exit. A single numeric argument is the k-point index (kp_read); then kp_info.dat is opened. Serial only (no MPI).

Input and output

Input

Common to both

• Standard input (interactive): Prompt “max k corrections, bin length, ibinit, iskip ?” → maxk, lbin, ibinitr, iskipr. If maxk < 0: yes_weight = .true. (replace weights by 1), maxk set to 0. If lbin < 0: lbin = -lbin, then “ngen max read ?” → iread. If ibinitr < 0: krep = iskip, ibinit = -ibinitr (multi-replica mode).

• fort.12: Used when iskipr ≥ 0. Unformatted. Record layout differs between readf and readffn (see Notes).

• fort.12.fn: Used when iskipr < 0; iskip = -iskipr - 1. Forward Walking output.

• Command line: First argument --help / -help / help, or a k-point index. Then kp_info.dat is read (first line: nkps; then k-point data). Only the chosen kp_read-th k-point records are used in the average.

readf — fort.12 record layout

• krep = 1, iskip ≠ 0: wbuf(j), wtot(j), eskip(1:iskip), ebuf(j,1).

• krep = 1, iskip = 0: wbuf(j), wtot(j), ebuf(j,1).

• krep > 1: wbuf(j), wtot(j), eskip(1), ebuf(j,1:iskip).

readffn — fort.12 record layout

• krep = 1, iskip > 0: wbuf(j), pip, wtot(j), eskip(1:iskip), ebuf(j,1).

• krep = 1, iskip = 0: wbuf(j), pip0, pip1. Then wtot(j) = 1 and ebuf(j,1) is set from wbuf, pip0, pip1 (or pip0/pip1 for sign) depending on optbra/iskip (Growth estimate).

• krep > 1: wbuf(j), pip, wtot(j), eskip(1), ebuf(j,1:iskip).

readffn only

• parminimized.d (required): First line skipped. Second line: nwr, optbra, gamma, tbra, etry. If optbra = -1 or -2, re-read with nsite. optbra = -2 → nodiv = .true. tbra, etry, nsite are used for Growth estimate output.

Output

• fort.20: First line “Independent bins nmis of length lbin”. Then for readf and readffn (when iskip ≠ 0 or tbra = 0): “Energy , error, # of bias correcting factor” and for k = 0,…,maxk: (gapjack, gaps), k. For readffn when iskip = 0 and tbra ≠ 0: “Growth estimate (H) error, # of bias correcting factor” with \(E_{\mathrm{try}} - \ln(\mathtt{gapjack})/t_{\mathrm{bra}}\) (or divided by nsite when optbra = -1) and its error.

• fort.21: One line per bin (ibin ≥ ibinit): ek(maxk,k)/wk(maxk) for k=1,…,krep, wk(maxk), ibin. Format suitable for block-averaged input (e.g. corrvar).

• Standard output: “number of measures done =” nmis; with k-point, “number of k-points / index read”, “kpoint/weight chosen:”, etc.

Notes

Jackknife and correction weights

Correction weight (bias factor k)

For each record the program reads wbuf, wtot, ebuf. It accumulates ek(k,:) and wk(k) using a “correction weight” wsk that depends on k = 0, …, maxk:

• k = 0: \(w_{\mathrm{sk}} = \mathtt{wtot}(j)\) (current measurement weight).

• k ≥ 1: \(w_{\mathrm{sk}}\) is wtot(j) times the product of wbuf from the previous k steps (j−1, j−2, …, j−k). When j−k falls in the previous block, wbufm (copy of previous block’s wbuf) is used.

So the k-th correction uses the product of the last k branching factors (wbuf) to reduce bias from population control. If maxk < 0, yes_weight sets wbuf and wtot to 1 (equal weights). In readffn with nodiv (optbra = -2), \(w_{\mathrm{sk}} = 1\) and wk is accumulated as \(w_{\mathrm{sk}} \times \mathtt{wtot}(j)\).

Per-bin weighted average (ekw, wkw)

Each bin has lbin measurements. When a bin completes, the accumulated ek(0:maxk,:) and wk(0:maxk) are stored in ekw and wkw. The bin-b average for correction k is:

\[\bar{E}_b^{(k)} = \frac{\mathtt{ek}(k)}{\mathtt{wk}(k)}\]

(final ek, wk for that bin). fort.21 writes the k = maxk value, wk(maxk), and bin index per line.

Jackknife formulas

The jack subroutine uses nbin bins:

1. Full weighted average (stored in gapjack, written to fort.20):

   \[\bar{E}^{(k)} = \frac{\sum_{b=1}^{n_{\mathrm{bin}}} \mathtt{ekw}(k,b,\cdot)}{\sum_b \mathtt{wkw}(k,b)} = \frac{\sum_b \mathtt{ekw}(k,b,\cdot)}{W^{(k)}}\]

2. Leave-one-bin-out Jackknife estimate (bin i excluded):

   \[\bar{E}_{(i)}^{(k)} = \frac{ W^{(k)} \bar{E}^{(k)} - \mathtt{ekw}(k,i,\cdot) }{ W^{(k)} - \mathtt{wkw}(k,i) }\]

   In code this is gap (per bin i).

3. Jackknife standard error (gaps):

   \[\sigma_{\mathrm{jack}}^{(k)} = \sqrt{ (n_{\mathrm{bin}}-1) \cdot \frac{1}{n_{\mathrm{bin}}} \sum_{i=1}^{n_{\mathrm{bin}}} \bigl( \bar{E}_{(i)}^{(k)} - \bar{E}_{\mathrm{jack}}^{(k)} \bigr)^2 }\]

   where \(\bar{E}_{\mathrm{jack}}^{(k)}\) is the mean of the \(\bar{E}_{(i)}^{(k)}\). fort.20 prints gapjack (average) and gaps (this standard error) for each k.

readf vs readffn

• parminimized.d: Read only by readffn; required for Growth estimate and nodiv.

• fort.12 record: readffn has pip (and pip0, pip1 when iskip = 0); readf has no pip.

• fort.20: readffn writes “Growth estimate (H)” when iskip = 0 and tbra ≠ 0; otherwise both write “Energy , error, # of bias correcting factor”.

• nodiv: readffn only (optbra = -2); weight accumulation differs.

• Typical source: readf for standard VMC/optimization fort.12; readffn for readforward fort.12.fn.

lbin, ibinit, maxk

• lbin: Generations per bin. nbin = countmax/lbin − ibinit + 1. Larger lbin → fewer bins, often smaller error but less resolution in fort.21.

• ibinit: Bins with index < ibinit are discarded (e.g. equilibration). Only bins with ibin ≥ ibinit contribute to ekw/wkw and Jackknife.

• maxk: Number of bias correction factors (k = 0,…,maxk). maxk = 0 gives plain weighted average; maxk > 0 adds corrections using products of previous wbuf. maxk < 0 at input → yes_weight and maxk set to 0.

k-points

• When the first command-line argument is a number, it is kp_read (k-point index). kp_info.dat must exist; first line is nkps. The program reads only the kp_read-th k-point’s records (or averages over kave_read consecutive k-points if kp_read > nkps). Record count in fort.12 must be a multiple of nkps (error 107 otherwise).

Related programs

• readforward.x: Produces fort.12.fn (Forward Walking). Use readffn to bin-average and get Growth estimate or energy with Jackknife error.

• corrvar.x: Reads fort.21 (and fort.22) for bootstrap of energy/variance. readf/readffn write fort.21 in a compatible block-averaged form.

• parminimized.d: Written by optimization or readforward; readffn reads optbra, tbra, etry, nsite from it.

• template/readf.input, template/readffn.input: Example stdin (one line: maxk, lbin, ibinit, iskipr).

Troubleshooting

Fatal errors (program stops)

• ERROR: file fort.12 not found or empty! (102) — readf or readffn with iskipr ≥ 0 could not open fort.12. Use fort.12 from the same run, or use iskipr < 0 and fort.12.fn.

• ERROR: file fort.12.fn not found or empty! (103) — iskipr < 0 but fort.12.fn missing. Provide fort.12.fn or run with iskipr ≥ 0 and fort.12.

• ERROR: file kp_info.dat not found or empty! (104) — First argument was a number (k-point index) but kp_info.dat is missing. Run without k-point index (or with 1) or create kp_info.dat (e.g. from readforward).

• ERROR: k-point index out of bounds! (106) — kp_read out of range. Use a valid k-point index (1 to nkps, or see code for kave_read).

• ERROR: number of measures not multiple of number of k-points! (107) — Record count in fort.12 is not divisible by nkps. Use fort.12 from the same k-point run.

• ERROR: wrong number of bins or k correcting factors! (108) — nbin ≤ 0 or maxk < 0 after applying yes_weight. Check lbin, ibinit, and that there are enough records; avoid invalid maxk.

Other notes

• fort.12 vs fort.12.fn format: Record layout differs between readf and readffn (readffn has pip, pip0, pip1). Use readffn for fort.12.fn from readforward.

• ibinitr < 0: Then krep = iskip and ibinit = -ibinitr; multiple replicas are averaged. Ensure iskip matches the file layout.